Market Power, Collusion, and Cartels II

Transcription

1 Market Power, Collusion, and Cartels II Paul T. Scott Toulouse School of Economics Empirical IO Spring / 58

2 Outline Genesove and Mullin (1998) Genesove and Mullin (2001) Albæk et al. (1997) Porter and Zona (1999) Asker (2010) 2 / 58

3 Genesove and Mullin (1998) "Testing static oligopoly models: conduct and cost in the sugar industry, " Genesove and Mullin (1998) 3 / 58

4 Genesove and Mullin (1998) Overview Bresnahan (1982) and BLP pioneered demand-based estimation of marginal cost and markups, but there are some concerns with these strategies: Functional form assumptions are crucial. θ might not be stable, in which case Bresnahan s regressions can give a biased estimate of the mean level of market power. Genesove and Mullin aim to test a Bresnahan-based estimation of costs for the sugar industry, where we have at least a rough idea of what marginal cost should be. Looking at the US sugar industry is interesting because the industry became more competitive; it was the time in between the Sherman Act s passage and when antitrust policy actually started being enforced. Also, price wars 4 / 58

5 Genesove and Mullin (1998) Industry background Sugar Trust controlled 80-95% of US sugar refining capacity in late 19th century There were documented periods of price wars in and following entries Dissolution of the trust in 1911 after federal government filed suit. 5 / 58

6 Genesove and Mullin (1998) Marginal costs: direct measures The main input in sugar refining is raw sugar, with approximately units of raw sugar needed per unit of refined sugar. A measure of refined sugar s marginal cost: c = c P RAW where c 0 represents the cost of inputs other than raw sugar. Genesove and Mullin argue that we can derive a lower bound on c 0 by assuming labor costs are fully fixed, and an upper bound by assuming labor is fully proportional to output. This places c 0 between 18 and 26, per 100 pounds of sugar. This is a small range of uncertainty as the non-raw-sugar inputs are only about 5% of costs. 6 / 58

7 Genesove and Mullin (1998) Identification of market power Recall Bresnahan s generalized pricing condition: P + θqp (Q) = c We can show that θ is equal to the elasticity-adjusted Lerner index: θ = η (P) P c P Thus, given demand estimates and a measure of cost, we can construct θ directly. However, we re also interested in estimating c and comparing to the direct measures. 7 / 58

8 Genesove and Mullin (1998) Demand GS consider a general demand function: Q (P) = β (α P) γ They estimate several versions of this demand system. For example, the estimating equation for the linear case (γ = 1) is: Q = β (α P) + ɛ. They use imports from Cuba to instrument for price (arguing that the only variable shifting Cuban imports are supply shocks in Cuba). 8 / 58

9 Genesove and Mullin (1998) Price wars 9 / 58

10 Genesove and Mullin (1998) Estimating θ After estimating demand, they can jointly estimate the cost parameters and θ. For the linear case, they estimate using the following moments: E [{(1 + θ) P αθ c 0 kp RAW } Z] = 0 Is the identification idea here the same as in Bresnahan (1982)? 10 / 58

11 Genesove and Mullin (1998) 11 / 58

12 Genesove and Mullin (1998) Dynamics and bias Note that the estimated θ is lower than the constructed θ. This might reflect bias resulting from dynamics (think back to Rotemberg and Saloner). 12 / 58

13 Genesove and Mullin (1998) Estimates and implied responses to a 68 increase in the raw sugar price. 13 / 58

14 Genesove and Mullin (1998) External validation One thing that can go wrong in the external validation is that misestimating k implies the wrong passthrough of inputs to costs: P = k P RAW The other thing that goes wrong is that if we have the wrong θ, we have the wrong passthrough of costs to price: P = θα + γc γ + θ For instance, the monopoly model predicts a price increase which is way too small because it predicts a very low cost-to-price passthrough. 14 / 58

15 Genesove and Mullin (1998) Comments Perhaps surprisingly, the sugar industry around 1900 appears to have been much closer to perfect competition than monopoly. The potential for bias from seasonality points to a broader issue: there s little reason to expect θ (or markups) to be stable in a changing environment. Therefore, one might say it makes more sense to use Bresnahan s strategy to validate a model of competition than as a reduced-form model on its own. 15 / 58

16 Genesove and Mullin (2001) "Rules, Communication, and Collusion: Narrative Evidence from the Sugar Institute Case" Genesove and Mullin (2001) 16 / 58

17 Genesove and Mullin (2001) Background In contrast to Sugar Trust (c ), the Sugar Institute (c ) was ostensibly a trade organization which was not explicitly aimed at limiting competition. Extensive internal memos reveal that it was undoubtedly unofficially aimed at limiting competition. The Institute served to help firms coordinate on rules which facilitated tacit collusion. In 1936, Supreme Court rules its practices illegal. "The stated aim of [the Institute s] rules was to eliminate discriminatory pricing... why it would have been in their interest to do so was never explained. The defendants... were silent on why compliance required collective action." 17 / 58

18 Genesove and Mullin (2001) Important features Some broad features are consistent with theoretical literature: Secret price cutting (understood broadly) was the main threat to cooperation. Collusion was sustained by threat of retaliation. Other features contrast with theories of collusion: Collusive agreements were incomplete (the games actual firms play are much more complicated than Bertrand or Cournot games). Extensive communication was involved; it s definitely not the case that firms only acquired information through some exogenous information structure. Cheating was typically not met with strong punishments (e.g., reversion to competitive conditions). Punishment strategies resembled tit-for-tat more than grim triggers. 18 / 58

19 Genesove and Mullin (2001) 19 / 58

20 Genesove and Mullin (2001) Secret price cuts "The Sugar Institute was primarily a mechanism to increase the probability of detection of sectret price cuts." But "secret price cuts" must be understood broadly. The Institute had many rules to avoid various forms of secret price cuts. The "full details" of sales of damaged sugar had to be documented. Favorable credit terms were banned as they are a substitute for price cuts. Refiners were prohibited from operating storage warehouses for customers through which discounts could be laundered. Refiners were required to enforce their contracts (especially specified delivery times) Freight rates could be cut rather than f.o.b. prices, and eventually refiners switched to c.i.f. (delivered) pricing. 20 / 58

21 Genesove and Mullin (2001) Quality suppression Some of the forms of secret price cuts could be understood as quality of auxiliary services, and the Institute s avoidance of them could be understood as collusion in quality suppression. "We view the supppression of non-price competition as complementary to contractual harmonization... If one is already choosing, and enforcing, one single contractual standard among many, one might as well limit nonprice competition along the way." 21 / 58

22 Genesove and Mullin (2001) Communication The first reason for extensive communication was in updating the terms of collusion: closing loopholes, updating to changing circumstances. This happened mostly at weekly meetings Firms also were expected to notify each other before many actions. This meant the firms knew what each other were up to, and if a firm was found to be engaging in an unapproved practice without notification, it would raise a red flag. Prior notification also facilitated mutually beneficial changes (e.g., if the monopoly price falls, all firms will want to lower their prices together) without triggering retaliation. The meetings were important to clarify when retaliation was warranted, and to ensure that retaliations were not seen as instances of cheating on their own. 22 / 58

23 Genesove and Mullin (2001) Punishments "When one firm openly lowered its rate for rail shipments... other firms would respond by lowering their rail rates to the same level. When the Pacific refiners gave a freight allowance on certain contracts, American announced that it would match it... the response to a deviation was generally restricted to the instrument of violation." These observations contrast with theories of optimal collusive equilibria in repeated games, where the best collusive equilibria involve the most extreme punishments available. 23 / 58

24 Albæk et al. (1997) "Government-Assisted Oligopoly Coordination? A Concrete Case" Albæk, Møllgaard, and Overgaard (1997) 24 / 58

25 Albæk et al. (1997) Abstract "In 1993 the Danish antitrust authority decided to gather and publish firm-specific transactions prices for two grades of ready-mixed concrete in three regions of Denmark. Following initial publication, average prices of reported grades increased by percent within one year. We investigate whether this was due to a business upturn and/or capacity constraints, but argue that these seem to have little explanatory power. We conclude that a better explanation is that publication of prices allowed firms to reduce the intensity of oligopoly price competition and, hence, led to increased prices contrary to the aim of the authority." 25 / 58

26 Albæk et al. (1997) Average concrete prices 26 / 58

27 Albæk et al. (1997) Prices at concrete plants around Aarhus 27 / 58

28 Porter and Zona (1999) "Ohio School Milk Markets: An Analysis of Bidding" Porter and Zona (1999) 28 / 58

29 Porter and Zona (1999) Overview Milk processors and distributors bid for school milk contracts on an annual basis. Unfortunately, the market is well suited to collusion. Price fixing convictions in 12 states with 90 convictions! Looking at auctions in the 1980 s in Ohio, Porter and Zona find that bidding behavior for most firms is consistent with competitive bidding, but behavior for accused firms is measurably different. 29 / 58

30 Porter and Zona (1999) The setting Demand is seen as very inelastic schools will pay a high price for milk if they have to. Milk is arguably a commodity, and firms bid only in price, so there is no incentive for product differentiation. Firms basically have the same production cost structure (milk processing is a mature industry), but delivery costs vary depending on plant and school locations. Firms typically face the same input (raw milk) costs due to regulation. 30 / 58

31 Porter and Zona (1999) Aspects facilitating collusion Bids and identities of bidders are publicly announced after auctions. Auctions are held at different times of the year for different schools. Multi-market contact (see Bernheim and Whinston (1990)) Milk processors are frequent customers of one another and have trade associtations. Typically a small number of plants are close enough to be viable suppliers for a given school. 45% of auctions receive one bid, 34% of auctions receive two, / 58

32 Porter and Zona (1999) Empirical model They estimate a model of bidder behavior with two pieces: A model of the probability firm j will submit a bid for the auction in school s A model of bid prices for submitted bids. Both models involve a bunch of characteristics of the firm, school, and (most importantly) the distance between the two. For non-accused firms, bid submissions and bid prices have the expected relationship with distance. 32 / 58

33 Porter and Zona (1999) 33 / 58

34 Porter and Zona (1999) 34 / 58

35 Porter and Zona (1999) On the other hand, firms in Cincinatti (which admitted to coordinating their bids for nearby schools) had relatively high bids for nearby auctions. 35 / 58

36 Porter and Zona (1999) 36 / 58

37 Porter and Zona (1999) Damages They do a reduced-form regression to assess damages. Basically, this involves regressing prices on the number of collusive firms involved in an auction. What are the limitations of this? What else could they do? 37 / 58

38 Asker (2010) "A Study of the Internal Organization of a Bidding Cartel" John Asker (2010) 38 / 58

39 Asker (2010) Overview A study of a bidding ring of stamp dealers, bidding on collectible stamps in New York auction houses. The ring used knockout auctions, internal auctions among members to allocate the good among ring members. The knockout mechanism leads to some interesting and counterintuitive effects: Side-payments provided incentives to bid above valuations. Overbidding sometimes caused inefficient allocations. Overbidding sometimes increased the price received by sellers. Overall, reduced competition more than compensated for the overbidding, and ring members benefited substantially from the scheme on average. 39 / 58

40 Asker (2010) The Knockout Auctions Before the actual (target) auction, ring members could submit bids in knockout auction run by a hired agent. The ring s bidding limit in the target auction is the maximum price from the knockout auction. A bidding agent would submit the ring s bid. If the ring wins the target auction, the highest bidder from the knockout auction gets the item and may owe side-payments to other knockout participants. "Sidepayments involve ring members sharing each increment between bids, provided that their bids are above the target auction price. Half the increment is kept by the winner of the knockout, and the balance is shared equally between those bidders who bid equal to or more than the "incremental" bid." 40 / 58

41 Asker (2010) 41 / 58

42 Asker (2010) 42 / 58

43 Asker (2010) Bidder heterogeneity 43 / 58

44 Asker (2010) Bidder D: "My objective, basically was, you know, make money from these people as opposed to actually buying the stamps." 44 / 58

45 Asker (2010) Naïve analysis Naïve estimates of damages can be easily calculated by assuming that knockout-auction bids represent true valuations. Then, the difference between the transaction prices in target auctions and the second highest bid in corresponding knockout auctions is a measure of damages (in cases where the second highest bid in the knockout was higher than the transaction price in the target auction). Note: target auctions were English auctions. However, incentives created by sidepayments call for a more careful assessment. 45 / 58

46 Asker (2010) 46 / 58

47 Asker (2010) Model basics Each bidder i has valuation in auction k of v ik [v i, v i ] drawn from F i (v). Valuations are private and independently distributed, but not identically distributed across bidders. Ring members know the number of other bidders participating in a knockout, but not the identities. 47 / 58

48 Asker (2010) Knockout bidding Expected profits: max b b (v ik x) h r (x) dxf i (φ (b)) where b b x (y x) h r (x) f i (φ (y)) dydx b (b x) h r (x) dx (1 F i (φ (b))) hr is the density function for the highest nonring bid, φ is the inverse strategy function, α j is the probability ( of j s participating in the auction, ) and F i (φ (b)) = j i α jf j (φ j (b)) / j i α j 48 / 58

49 Asker (2010) Optimal bidding FOC for profit maximization: (v ik b) h r (b) F i (φ (b)) + b (v ik x) h r (x) dxf i (φ (b)) b (b x) h r (x) dxf i (φ (b)) + 1 b 2 h r (x) dx (1 F i (φ (b))) 49 / 58

50 Asker (2010) Recovering valuations The first-order condition cannot be inverted for v in general, but with only two bidders, v ik = b 1 2 H r (b) (1 G i (b)) (h r (b) G i (b) + H r (b) g i (b)) where G i is the distribution function of b i. Asker focuses on auctions with two bidders to avoid identification issues. 50 / 58

51 Asker (2010) Overbidding Lemma 1 states that π ik b ik bik =v ik 0. Therefore, knockout bids are weakly greater than valuations. Corollary: the knockout auctions can lead to inefficient allocations. 51 / 58

52 Asker (2010) Auction heterogeneity Extending the model to allow for unobserved auction-level heterogeneity, write valuations as: u ik = e x kγ (v ik ε k ). Asker s structural approach recovers the distribution of v s and ε s. We re going to ignore details of dealing with the ε s here, but you should be able to see how the distribution of v s could be estimated if we don t have the ε s (think GPV). 52 / 58

53 Asker (2010) Bidder heterogeneity For simplicity, he classifies bidders as either "weak" or "strong" and estimates a different distribution of valuations F ( ) for each type. Remember that bidders don t know which other bidders are participating. Empirical frequencies of each bidder s participation are used for α j s. 53 / 58

54 Asker (2010) Notes on counterfactuals Solving for equilibria of the knockout auctions might be hard, but his counterfactuals are only English auctions, which are analogous to second price auctions and therefore easy to solve. This makes counterfactuals WAY easier. A difficulty is not knowing the distribution of (second highest) nonring bids. U.B. assumption: second highest nonring value is equal to highest nonring valuation. This provides upper bound to damages. Why? L.B. assumption: second highest nonring value is equal to minimum of highest nonring valuation and highest ring valuation. The provides lower bound to damages. Why? 54 / 58

55 Asker (2010) 55 / 58

56 Asker (2010) 56 / 58

57 Asker (2010) 57 / 58

58 Asker (2010) 58 / 58